Multi-thread implementation of a fuzzy neural network for automatic ECG arrhythmia detection

Abstract

A fuzzy neural network was implemented using a multithreading approach for detection of atrial fibrillation, bigeminy, and normal sinus rhythm in the MIT-BIH Arrhythmia Database. The feedforward multilayer perceptron neural network produces fuzzy outputs due to a modification of the learning algorithm that changes the crisp target labels for fuzzy target labels. The input data to the neural network consisted of nine inputs: Seven contiguous RR intervals, their average and their standard deviation. The trained fuzzy neural network was implemented using concurrent thread synchronization with critical sections for mutual exclusion and process synchronization with semaphores. Concurrent process synchronisation is slower but allows data sharing among different process. Sensitivity and positive predictivity rates above 90% for atrial fibrillation episode and duration detection were reached in the database.